Do AI systems think fast or slow? Testing dual-process patterns in AI risk assessment

This study investigates whether AI systems exhibit cognitive patterns similar to human dual-process thinking by testing AI risk assessment across varying information conditions. Using ChatGPT-4o, we evaluated risk ranking capabilities under five conditions from minimal guidance to structured analytical frameworks, comparing results against human cognitive benchmarks. Results demonstrate that AI operates in a hybrid default state, showing moderate correlations with both intuitive and analytical human benchmarks under minimal guidance. Structured information design enhanced AI’s analytical capabilities, with correlations reaching 0.768 with expert deliberations and 0.921 with quantitative models. AI showed systematic differences from human risk prioritization, demonstrating moderation rather than extreme responses characteristic of human intuitive assessment. These findings indicate that AI can serve as a valuable decision-support tool for systematic risk assessment when properly configured but operates through fundamentally different cognitive patterns than humans.